Why data-driven attribution (DDA) is the most effective model

By Shannon Reeves

 

Google making data-driven attribution (DDA) its default attribution model may have left you asking a few questions… does my account have enough data to utilise DDA effectively? How will this affect the performance of my campaigns? How does DDA work, anyway?

Well, I’m going to discuss how different attribution models work; why DDA is one of the most effective models out there; and whether you can start using this model for more accurate reporting on your marketing activity right away.

What is an attribution model?

Put simply, attribution uses statistical analysis to determine how credit is awarded to individual touch points along the path to conversion.

An attribution model provides different ways of allocating credit to those touch points, based on their importance in the customer journey. Each model measures the importance of the touch points differently by using different analytical techniques.

The insight provided by attribution models allows us to determine which combination of channels or activity is the most effective, and therefore optimise our strategies accordingly, and so access to the most accurate data is extremely valuable.

The benefits of using an attribution model

1. Valuable insights

By analysing attribution, we’re able to determine the best performing channels, campaigns, creatives and other key elements to delivering conversions.

This allows us to optimise accordingly as well solving any potential problems.

2. Optimised spend

By understanding which touch points are delivering the strongest performance, we are able to up or down weight budgets accordingly for optimal results.

3. Improved personalisation

Attribution data can be used to understand the channels and messaging preferred by users, allowing for more effective targeting and personalisation.

We can hone messaging and creative elements, as well as better understand how and when to communicate with users.

4. Improved performance

Being able to reach the right user with the right message at the right time can lead to increased conversion rates and higher ROI.

5. Customer LTV optimisation

Attribution models can show the customer lifetime value influence of each touch point, which helps us to determine spend across prospecting and retargeting campaigns.

Challenges & mistakes when using an attribution model

Whilst attribution analysis offers many benefits, there are some common challenges and mistakes that can result in misattribution, obscuring the success of campaigns:

1. Correlation-based bias

Attribution models can be subject to correlation-based biases when analysing the customer journey, making it look like one event has caused another when it may not have.

2. Cheap inventory bias

Lower cost media may appear to perform better due to the natural conversion rate for the targeted user.

3. Brand & behaviour

Attribution models can often overlook the relationship between brand perception and consumer behaviour. They should be able to detect relationships between brand building initiatives and conversions.

4. In-market bias

This refers to in-market consumers who would have converted organically had they not been served the ad. However, the ad gets the attribution.

5. Digital signal bias

This occurs when attribution models do not factor in the relationship between online activity and offline sales.

6. Missing message signal

A common mistake is evaluating creative in aggregate and determining that a message is ineffective when, in reality, it would be effective for a smaller, more targeted audience.

7. Data wrangling

No attribution model can provide full visibility into the customer journey. Therefore, we sometimes have to employ multiple models and correlate the data from each to gain the most accurate insights.

The volume of data and the complexity of the models can pose a challenge, with many marketers spending more time aggregating the information in a digestible way than deriving meaningful insights from it.

8. Lack of visibility into external factors

Without incorporating aggregate information, we do not have visibility into external trends that might be affecting marketing efforts, such as seasonality.

What are the different attribution models?

1. Single-touch attribution models

Single-touch attribution models assign all credit to one touch point. They fail to factor in the broader customer journey and therefore alter the perceived effectiveness of other channels. For these reasons, they shouldn’t be relied upon completely.

• First-click

This model gives full credit to the first touch point the user interacted with, regardless of the outcome or any additional messaging seen subsequently.

This works for marketers who are solely concerned with lead forms and demand generation and is good for showing how top-of-the-funnel marketing techniques generated initial awareness, eventually leading to a conversion.

It is a good fit for short buying cycles and easy to implement. However, it over-emphasises a single part of the customer journey and, since it doesn’t reveal what actually drove a conversion, doesn’t allow for optimisation.

• Last-click

Conversely, the last-click model gives full attribution credit to the last touch point the user interacted with before converting, without accounting for prior engagements.

This is the default attribution model in most platforms and works for marketers who are solely focused on driving conversions and don’t value non-converting factors.

Like first-click, it is a good fit for short buying cycles and is easy to implement. However, it ignores the various influences that affect a customer’s path to purchase and solely focuses on the last interaction.

This causes problems when it comes to correctly reporting conversions because every other contributing channel is ignored.

2. Multi-touch attribution models

Multi-touch attribution models award each contributing touch point credit for the final conversion. As a result, these are considered appropriate for more situations. These models are differentiated by how they divide credit between touch points.

• Linear

This model records each touch point engaged with that leads to conversion. It weighs each of these interactions equally, giving each touch point the same amount of credit towards driving the conversion.

This modelling system allows us to optimise the entire customer journey as opposed to just one touch point and demonstrates that each channel has value in a nuanced and straightforward manner.

However, because it assigns the same value of credit to both high-performing and low-performing touch points, it is better suited to delivering a holistic strategy as opposed to optimising based on performance.

• Position based

Unlike linear attribution, the position-based model scores engagements separately, noting that some are more impactful than others on the path to purchase.

Specifically, this follows a U-shape, with, for example, both the first and last touch points are given 40% of the credit. The remaining 20% is divided amongst the touch points engaged with in between.

This is beneficial because every touch point in the customer journey receives some credit, whilst still prioritising the first and last touch points.

This allows us to optimise the touch point that introduced the customer to the brand and the touch point that converted the customer.

However, it may be prone to assigning inaccurate value to the first and last touch points.

• W-Shaped

This uses the same idea as the position-based model. However, it includes one more core touch point: opportunity creation.

As an example here, the touch points credited with first-touch, last-touch, and opportunity creation each receive 30% of the credit. The remaining 10% is divided amongst the additional engagements.

This is a good model to use when there is a clearly identifiable “opportunity creation” stage taking place.

However, it’s not always easy to neatly categorise stages of the user’s journey.

• Full path

This is a highly technical and sophisticated model. It builds on the W-Shaped model, incorporating an additional touch point: lead creation.

This describes the touch point when a user became a qualified lead.

The bulk of the credit (25%) is given to the major milestones of the customer journey (first-touch, lead creation, opportunity creation, and last-touch), with the remaining 10% divided amongst the additional touch points in between.

This model works well for B2B marketing as well as for high value purchases that require a great deal of customer consideration as it provides a granular view of the customer journey.

However, it doesn’t lend itself well to campaigns that focus on low-involvement purchases.

• Time decay

The time decay model also weighs each touch point differently on the path to purchase.

However, this model gives the touch points engaged with closer to the conversion more credit than those engaged with early on, assuming that those had a greater impact.

This allows us to optimise the touch points that drove conversions as well as the touch points that assisted.

It is a helpful way to conceptualise relationship-building and is a good fit for long buying cycles.

However, it places little value on the touch point that drove a customer to a business in the first place; minimising the effect of top-of-the-funnel marketing techniques and ignoring valuable touch points that happened earlier in the customer journey.

• Custom

If you want to assign your own attribution weights to each touch point (based on industry, channels used, typical buyer behaviour, or simply what you deem to be most important), you can create a custom attribution model.

This has the potential to give a more accurate representation of the impact each touch point has on the customer journey and is a good fit for longer buying cycles.

However, this approach can be difficult and time-consuming to implement and requires a lot of data.

What Is data-driven attribution (DDA) and why is it the most accurate?

Data-driven attribution (DDA) works differently to the above-mentioned attribution models.

Unlike the time-consuming nature of the custom model, it uses machine learning to automatically analyse the relevant data and calculate the actual contribution of each touch point along the path to conversion.

By comparing the paths of converting and non-converting customers, DDA determines the true value within the conversion path, discovering the interaction patterns that lead to conversion. Credit is awarded based on the true incremental value of each individual touch point, resulting in better optimised campaigns and more accurate reporting.

This rigorous and easy-to-use attribution model can result in better campaign performance by allowing for time-efficient and data-driven optimisations – especially when combined with automated bidding strategies.

The model will continue to optimise and improve over time with the more data it gathers.

Can I use DDA on my PPC campaigns?

It used to be the case that DDA was only available for specific conversion actions and accounts with enough conversions in their recent history.

Now, however, the data requirements for conversion history have been removed and support is being added for more conversion types.

DDA has now replaced last-click ad as the default model in Google Ads – so, there should be no reason why you can’t make the switch and improve the accuracy of your attribution right away!

To do this, simply follow the instructions below:

1. In Google Ads, go to Reporting & Attribution > Attribution.

 

2. Choose a Floodlight configuration. All impressions and clicks tracked through this configuration will be taken into consideration by the model.

 

3. Click Attribution Modelling Tool.

 

4. Click dropdown next to the default model, at the bottom of the list click ‘Create new data-driven model’.

 

5. Name your model. Tip: Include today’s date.

 

6. To train the data-driven model correctly, select all Floodlight activities that represent conversions. (Don’t include Floodlight activities that were created solely for building audience lists; such as remarketing pixels.) Filter reports to a single Floodlight activity later.

 

7. Choose Basic Channel Grouping (for most setups).
   1. To create a custom channel grouping go to: Reporting & Attribution > Attribution > Attribution Modelling Tool > Channel Groupings.

 

8. (Optional) Set a custom lookback window.

 

9. Click Save.

NB: You can only create one data-driven attribution model. Your model needs to train for 1 to 2 days and will be greyed out until usable.

If you make changes to an existing model, the model will need to retrain, which may take up to 9 days. Reporting is unavailable for date ranges prior to a model being trained.

Making the most of DDA within PPC

Below are four key tips to help you make the most of your DDA model:

1. Switch to an automated bidding strategy

By combining a DDA model with automated bidding strategies, Google can predict the incremental impact each specific campaign element will have on driving a conversion and adjust bids accordingly to maximise performance.

2. Optimise your bids

Switching to a new attribution model will change all associated conversion metrics and your bids will need to be updated to reflect this new data. If you switch to DDA but keep using bids tailored to the last-click model, you may end up over or under bidding resulting in poor efficiency and performance.

3. Allow some time

Because DDA works by measuring multiple clicks within a conversion path, it may take some time before accurate attribution can be reported back.

You will therefore need to allow time for users to complete their conversion path before evaluating results.

It is worth giving the model at least a couple of weeks after switching over to gather this data.

4. Re-evaluate your data

Users tend to refine their searches as they get further down the path to conversion.

This means that, when running with a last-click attribution model, high-intent keywords may have been awarded full credit for conversions that originated from broader, more generic terms earlier in the user journey.

Now that your attribution is more accurate, it is worth re-evaluating which campaign elements are truly driving the best performance and optimising accordingly. For example, you might start to see that generic campaigns are actually responsible for more conversions than your brand campaigns were giving them credit for.

We hope you have found this article helpful. If you have any burning attribution-related questions or need some help with optimising and reporting on your own accounts, please get in touch for a chat.